Quota

A quota is an upper limit on the use of a service. It defines how much a single customer is allowed to consume within a certain period of time. This can be a number of requests per minute, an amount of storage space, or a sum of money per month. Once the limit is reached, the provider rejects further requests or throttles them. The term comes from Latin and roughly means “share”. That is exactly what it describes: the share of a shared resource that you are entitled to.

Why providers have to set limits

A data center has a finite amount of hardware. When thousands of customers access it at the same time, the computing power has to be divided up somehow. Without upper limits, a single user could bring the service to a halt for everyone else. Sometimes this doesn’t even happen intentionally: a faulty program stuck in an endless loop fires off millions of requests per hour.

The second reason is money. With AI services, every single response costs the provider real computing time, and that gets billed. A quota therefore also protects the customer from themselves. Anyone who sets a spending limit of 50 euros a month won’t get a nasty surprise if their program has a bug.

Third, quotas are a sales tool. The free tier of a service has tight limits, the paid tiers have higher ones. Whoever needs more, pays more. For companies, this becomes a problem when their product suddenly gets popular: user numbers grow faster than the provider raises the limit.

How the limit is monitored technically

The provider keeps count. Every request is attributed to your account and added to a counter. If the counter exceeds the allowed number, an error message comes back instead of a response. On the web, this message has a fixed number: error 429, “Too Many Requests”. A well-written program then waits briefly and tries again.

A common method is called the token bucket. Picture a bucket that constantly has drops of water trickling in, say ten per second. Every request takes one drop out. If the bucket is empty, you have to wait. The advantage: anyone who has paused for a while is briefly allowed to send a surge of requests, because their bucket is full. This is called a burst.

It’s important to distinguish this from the related term rate limit. A rate limit governs the speed, i.e. requests per second or minute. A quota, in the narrower sense, governs the total amount over a longer period, such as a month. Many providers use both at the same time and use the words interchangeably in everyday speech.

Quotas in ChatGPT, cloud services, and at school

Quotas are most visible with chatbots. Free accounts are only allowed to use the best model a limited number of times per day. After that, it switches to a weaker model, or you have to wait until the next day. Image generators work the same way and bill in images per month.

In the cloud, meaning on rented servers on the internet, quotas are ubiquitous. New customers there are often only allowed to book a handful of graphics cards at the same time. Anyone who wants more has to apply for an increase and justify it. This topic comes up in business news when providers lower limits due to chip shortages – a clear sign that demand exceeds supply.

You know this principle outside the AI world too. Storage space in the school cloud is limited, as is the data volume of your phone plan. A common misconception is to mistake a reached quota for a technical malfunction. In fact, the system is working exactly as intended – it’s just telling you no right now.